Figure 6
From: A new periocular dataset collected by mobile devices in unconstrained scenarios
Pairwise filters CNN architecture. This model contains filters that directly learn the similarity between a pair of images. The output informs whether the images are of the same person or not.
